Incident sheet · /security
- System
- Security & Risk
- Filed
- Read
- 4 min · 765 words
- Severity
- Routine
Leaked Cloud Keys in Docker Hub Images Get Stolen in Under 8 Seconds
Honeypot images with planted AWS, OpenAI and Anthropic keys were scraped within seconds — one AWS key was used 7.86 seconds after publication on Docker Hub, and all 100 keys within 15 minutes.

Event log
- A planted AWS key was exploited 7.86 seconds after its image was published on Docker Hub; median first-use time was 51.19 seconds.
- All 100 planted AWS keys were used within 15 minutes, receiving over 140,000 API calls from 5,000 unique IP addresses.
- 300 honeypot images across eight public registries were pulled more than 75,000 times over four months.
- Scanners extracted secrets via static analysis of image layers without running the containers.
- Docker acknowledged the findings in May but implemented no changes as of an August follow-up test.
Attackers exploited a planted AWS key just 7.86 seconds after the container image carrying it went live on Docker Hub. That number comes from a four-month honeypot experiment by two Stony Brook University researchers, Billy Tsouvalas and Nick Nikiforakis, presented yesterday at the ACM Conference on Computer and Communications Security (CCS). For any team that publishes container images, the finding sets a hard clock on how long a leaked secret stays secret: not minutes, seconds.
The two researchers published 300 honeypot images across eight public container registries, including Docker Hub and Harbor. Inside those images they placed live-looking AWS, OpenAI, and Anthropic API keys, plus decoy SSH and database credentials. They then instrumented the images to alert them whenever someone built one, ran one, or opened a shell inside it — and they watched what happened to the keys on the providers' side.
A lot happened. The honeypot images were pulled more than 75,000 times. The planted AWS keys received over 140,000 API calls from roughly 5,000 unique IP addresses. The takeaway: public registries are under continuous, automated surveillance the moment an image lands.
How fast do the attackers actually move?
The speed was the most striking result of the study.
- The fastest observed use of a planted AWS key came 7.86 seconds after publication on Docker Hub.
- The median time to the first API call was 51.19 seconds.
- All 100 planted AWS keys were used within 15 minutes.
To put that in practical terms: a developer who pushes an image with a hardcoded key does not have a coffee break's worth of margin. By the time a code review or a CI pipeline finishes, the key is likely already in someone else's hands.
The AWS traffic followed a familiar escalation pattern. The scanners first checked whether a key worked, then listed the resources it could reach, and in some cases went on to consume them. The OpenAI and Anthropic keys fared no better — the attackers hijacked them quickly and drained the test-account balances dry.
Docker Hub accounted for nearly all of the AWS key abuse in the experiment. Harbor, one of the other seven registries tested, also saw some AWS activity. In short, the exposure is not evenly distributed; it concentrates where the scanning is easiest and the traffic is largest.
Who is doing the stealing?
The researchers gave the automated scanners a collective name — "ContainerBots" — but they caution that the label is loose. The scan traffic showed no evidence of coordination by a single operator. This is not one adversary; it is an ecosystem of independent tooling all pointed at the same firehose of freshly published images.
The mechanism behind the thefts surprised the researchers. Their monitoring almost never caught anyone actually running the honeypot containers. Instead, the scanners extracted credentials by inspecting the image layers directly, without ever instantiating the container.
"The scanners interacting with our images overwhelmingly do so through static analysis, e.g., unpacking image layers or querying the registry API to extract filesystem contents, without ever instantiating the container," the researchers explain.
That detail matters for defenders. Runtime protections, sandboxing, and behavioral monitoring inside the container never get a chance to fire, because the attack happens on the artifact, not the workload. The image layer is the attack surface.
What does this mean in practice?
This class of leak is common in the real world, and recovering from one is expensive. Revoking a leaked key means finding every system that uses it, rotating credentials in a coordinated way, and auditing what the key touched before the rotation — work that grows with the size of the estate.
The researchers put forward two concrete fixes for the platform side:
- Docker Hub should scan images for embedded secrets before pushing them live, and warn publishers when secrets are detected.
- Access to Docker Hub's public feed of recently published or updated images should be restricted or rate-limited, to blunt mass-scanning tools that feed on it.
Has Docker responded?
Not visibly. Tsouvalas told BleepingComputer that the researchers reported their findings to Docker in May, along with the full paper. Beyond an acknowledgment, they received no follow-up replies. A small-scale experiment they ran in August to check whether any defenses had been implemented on the platform found no changes.
For enterprises, the lesson does not depend on Docker acting. Treat every secret that has ever touched a container image as compromised the moment that image is public, and build rotation and scanning into the pipeline rather than the incident response.
via bleepingcomputer.com (Source)
More from Amara Osei
Show full bio
Correspondent covering regulation, policy and their impact on enterprise information technology and the leaders who run at CIO Monitor.
188 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Operation ShadowCloud: Backdoors Hide C2 in GitHub, SharePoint and Ably
- Barracuda Cloud Breach Traced to St Vendor Key and AI Pentesting Tool
- Railway Breach Cost $740,000 in Fraudulent AI Usage
- US Seizes Infrastructure Behind $400M North Korean Crypto Theft Ring
- Iris Ohyama cloud breach may affect 7.7 million people